Create video from image, video, and audio references with MiniMax H3 Max on Muapi. Up to 1080p, pay per generation. MiniMax H3 Max Reference to Video creates high-fidelity video from a prompt plus image, video, and optional audio references through the Muapi API.
MiniMax H3 Max Reference to Video creates high-fidelity video from a prompt plus image, video, and optional audio references through the Muapi API.
このモデルについて
MiniMax H3 Max Reference to Video generates a video clip guided by a prompt plus up to 12 combined reference images and videos, with optional reference audio layered in. Reference the assets by name in your prompt (for example "Image 1" or "Video 1") to tell the model how each one should influence the result — a character's appearance, a background style, a motion pattern, or a soundtrack. Output resolution runs up to 1080p across durations of 5-15 seconds, with an adaptive aspect ratio option that follows the references automatically. It follows Muapi's standard asynchronous request/poll/webhook pattern. For a single starting image without multi-reference blending, see MiniMax H3 Max Image to Video; for a pure text prompt with no references, see MiniMax H3 Max Text to Video.
コスト分析
| プロバイダー | 費用 | 備考 |
|---|---|---|
| muapiapp | 開始価格: $0.065/秒 (480p) up to $0.208/秒 (1080p), plus reference item fees | Per-second output cost by resolution; the first 4 reference items are included, each additional reference image/video/audio item adds a small flat fee. 生成ごとの従量課金, サブスクリプション不要. |
| Replicate | 利用不可 | MiniMax H3 Max is 掲載なし Replicate. |
Per-second output cost by resolution; the first 4 reference items are included, each additional reference image/video/audio item adds a small flat fee. 生成ごとの従量課金, サブスクリプション不要.
MiniMax H3 Max is 掲載なし Replicate.
** 競合サービスの料金は類似のモデル構成および利用ティアに基づいて算出された推定値です。
設定スキーマ
| パラメータ | 型 | 説明 | デフォルト |
|---|---|---|---|
| プロンプト | string | Describe the video scene and how the references should be used (e.g. 'Image 1', 'Video 1'). | Use the references to create a cinematic product reveal with a slow camera orbit. |
| Reference 画像s | array | Reference image URLs. Provide at least one image or video reference. Max 12 combined reference files. | undefined |
| Reference 動画s | array | Reference video URLs (2-15s each, 15s combined max). Provide at least one image or video reference. | undefined |
| 参照音声 | array | Optional reference audio URLs (2-15s each, 15s combined max). Audio cannot be used alone. | undefined |
| アスペクト比 | Enum（7個の選択肢） | - | adaptive |
| 解像度 | Enum（3個の選択肢） | - | 768p |
| 長さ（秒） | Enum（11個の選択肢） | 生成する動画または音声の長さ（秒単位）。 | 5 |
| Prompt Expansion Mode | Enum（3個の選択肢） | Prompt rewriting effort. 'quality' can add up to ~30s of processing. | balanced |
Describe the video scene and how the references should be used (e.g. 'Image 1', 'Video 1').
Use the references to create a cinematic product reveal with a slow camera orbit.Reference image URLs. Provide at least one image or video reference. Max 12 combined reference files.
undefinedReference video URLs (2-15s each, 15s combined max). Provide at least one image or video reference.
undefinedOptional reference audio URLs (2-15s each, 15s combined max). Audio cannot be used alone.
undefined-
adaptive-
768p生成する動画または音声の長さ（秒単位）。
5Prompt rewriting effort. 'quality' can add up to ~30s of processing.
balanced開発者ドキュメント
Add references: Provide reference_images and/or reference_videos URLs (at least one of the two is required, up to 12 combined files); reference_audios is optional and cannot be used alone.
Write the prompt: Describe the scene and explicitly reference your assets by name, e.g. "Use Image 1's character walking through the scene from Video 1."
Pick resolution and duration: Set resolution to 480p, 768p, or 1080p, and duration to any value from 5 to 15 seconds.
Set aspect ratio: Leave aspect_ratio as adaptive to follow the references automatically, or choose a fixed ratio.
Submit and retrieve: POST the request, then poll /predictions/{request_id}/result or supply a webhook_url to receive the finished video URL.
curl -X POST https://api.muapi.ai/api/v1/minimax-h3-max-reference-to-video \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Use the references to create a cinematic product reveal with a slow camera orbit.",
"reference_images": ["https://example.com/product.jpg"],
"resolution": "1080p",
"duration": 6
}'
FAQ
It generates a video guided by a prompt plus one or more reference images, videos, and optional audio, blending their appearance, motion, or sound into the new clip.
Up to 12 combined reference images and videos, plus optional reference audio tracks up to 15 seconds combined.
No, reference audio must be paired with at least one reference image or video.
Resolution can be 480p, 768p, or 1080p, and duration can be any value from 5 to 15 seconds.
When `aspect_ratio` is left as `adaptive`, the output follows the aspect ratio implied by your reference assets instead of a fixed value.
Poll the request ID returned by the initial POST at `/predictions/{request_id}/result`, or pass a `webhook_url` to be notified when the video is ready.